Authors' reply

In their thoughtful comments about our study, Cella & Preti raised several important points about some of the current challenges and limitations of research synthesis methods. ¹

In systematic reviews, search strategies need to strike a balance between specificity with regard to the research question at hand and sensitivity to capture the broadest number of relevant studies. Given that our only outcome of interest was negative symptoms in psychotic disorders, we included the term ‘negative symptom’ in our search strategy, as well as 26 synonyms of negative symptoms and associated terms, and broad keywords for psychosocial interventions. Our search retrieved a comprehensive 4136 non-duplicate studies from five major databases. Although some studies might have been missed, their omission from our review is unlikely to have been systematic, and our findings are overall consistent with those from a previous review. ² We do agree with Cella & Preti that publication bias could be a problem – it is a common threat to almost all reviews – and we found some evidence of it, which we reported in the paper. We also acknowledge that the general intervention categories in our review might not perfectly match the various psychosocial interventions that have been tested in the literature. Our goal was to provide readers with a broad sense of the benefits of different intervention approaches for negative symptoms. Care was given to fully present the characteristics and quality of each study in the paper and online supplement.

Cella & Preti pointed out that our meta-analysis could have focused on change scores rather than endpoint scores of negative symptoms to account for baseline differences in groups. This approach may be more efficient under certain circumstances. In randomised trials, however, both approaches can be assumed to provide the same underlying intervention effects if the trials are adequately randomised: the difference in mean endpoint scores will be the same on average as the difference in mean change scores. To our knowledge, the decision for using one measure over another is currently not resolved, with evidence suggesting that endpoint scores tend to produce more conservative estimates. ³ Cella & Preti further suggested applying a restricted maximum likelihood estimation method to the meta-analysis. Research synthesis methodology is a developing field and there is no consensus at present on the best approach for random-effects meta-analysis. Although valuable alternatives to random-effects estimation methods have been developed, ⁴ DerSimonian-Laird remains the most widely used approach and an adequate method in most scenarios. ⁵

Finally, Cella & Preti raised two points about the way control conditions were handled in the review. We would like to reassure readers that we did not double-count studies by considering active control conditions for one analysis as active treatment conditions in another analysis. We further wish to clarify that treatment participants in all trials received the psychosocial intervention as an adjunct to treatment as usual (TAU), as doing otherwise would be considered unethical.